Perception Test 2025: Challenge Summary and a Unified VQA Extension
Joseph Heyward, Nikhil Pathasarathy, Tyler Zhu, Aravindh Mahendran, João Carreira, Dima Damen, Andrew Zisserman, Viorica Pătrăucean · 2026
Read Perception Test 2025: Challenge Summary and a Unified VQA Extension by Joseph Heyward, Nikhil Pathasarathy, Tyler Zhu, Aravindh Mahendran, João Carreira, Dima Damen, Andrew Zisserman, Viorica Pătrăucean (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-06287.txt.
Description
Joseph Heyward and 7 co-authors summarize the third Perception Test challenge in a 2026 report. Held as a workshop at a major computer vision conference, the challenge benchmarks video models on multimodal perception. Two guest tracks covered image understanding and a video generation physics test.
This edition emphasized unification: competitors had to use single approaches rather than task-specific pipelines. Five consolidated tracks covered video question answering, merged object and point tracking, merged action and sound localization, grounded video QA, and hour-long video QA. One track recast classic perception tasks as multiple-choice video questions that video-language models can answer natively. The results highlight how hard diverse perception remains for current models under unified interfaces.
Catalog record
| CREATOR | Joseph Heyward, Nikhil Pathasarathy, Tyler Zhu, Aravindh Mahendran, João Carreira, Dima Damen, Andrew Zisserman, Viorica Pătrăucean |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-06287